The study verifies the usefulness of selected machine learning techniques for predicting hourly demand for electricity within a short time period. The results of the performed analyses show that the lowest values for both the MAPE forecast error for the test set at the level of 17% and the lowest share of the balancing energy in the total consumption at a level which does not exceed 15% were obtained for models for which the input data included the averaged electricity consumption profile for characteristic days of the week, the forecast number of pure production pieces and the encoded day of the week and time of the day. Among the tested models, forecasts prepared on the basis of artificial neural networks and standard CRT trees were characterised by the best quality of predictions.